Solving petrological problems through machine learning: the study case of tectonic discrimination using geochemical and isotopic data